Professional Experience

Design and build production-grade Agentic AI systems for legal research, focusing on accuracy, traceability, and scalability. Architect multi-agent orchestration combining LLM planning with deterministic guardrails for reliable legal AI.

• Agentic AI System Architecture (Legal Research):
  • Architect multi-agent orchestration for legal research using LangGraph, integrating RAG and specialized tools via MCP to retrieve and synthesize evidence from diverse sources.
  • Implement hybrid workflows where LLMs plan and reason while deterministic services enforce validation, tool access, state transitions, stopping criteria, and output contracts.
  • Engineer reliable streaming experiences using Server-Sent Events (SSE) with typed errors, timeouts, cancellation, and comprehensive observability.
  • Optimize model routing across GPT, Claude, and Gemini based on capability, quality, risk, latency, and cost with explicit fallback and escalation policies.
  Tech: LangGraph, RAG, MCP, Multi-Agent Systems, SSE, GPT, Claude, Gemini, Prompt Engineering, Observability.
• LLM Evaluation & Quality Frameworks:
  • Design evaluation frameworks measuring hallucination, misattribution, relevance, completeness, source coverage, and citation integrity.
  • Improve grounding through query decomposition, retrieval optimization, reranking, provenance tracking, citation reconciliation, and structured generation.
  • Lead controlled experiments across prompts, retrieval strategies, model configurations, and generation parameters; validate schemas and track provenance before reporting.
  • Partner with product, engineering, and research teams to translate requirements into measurable AI capabilities.
  Tech: LLM Evaluation, RAG, Grounding, Citation Integrity, Query Decomposition, Reranking, Provenance Tracking, Structured Generation.

As a Senior Data Scientist (Agentic AI) at Guardian Life, led initiatives to automate and optimize RFP (Request for Proposal) documents using advanced AI techniques, significantly improving data extraction accuracy and operational efficiency through systematic prompt engineering and evaluation strategies.

• RFP Documentation Automation System (Agentic AI):
  • Spearheaded the development of an Agentic AI system to automate RFP documentation processes, reducing manual effort by 60%.
  • Improved prompt engineering through comprehensive analysis of RFP documentation structures using few-shot learning and chain-of-thought prompting.
  • Developed thorough documentation of the RFP AI project including system architecture, data flows, and integration specifications.
  • Implemented comprehensive logging to track extraction accuracy and system performance metrics.
  • Enhanced confidence score calculations by refining logic to better reflect extraction certainty.
  • Applied advanced AI knowledge by configuring LLM parameters strategically (deterministic for extraction, creative for analytical components).
  • Increased accuracy of marketing contact extraction by 35% and boosted confidence scores by 28%.
  Tech: Amazon AWS EC2, Bedrock, boto3, Claude Sonnet, Prompt Engineering, LLM Configuration, loguru, logging.

Worked in end-to-end development of AI projects, mentored team members, and built models for forecasting, recommendation, and adverse event detection using patient and drug data.

• AI Chatbot for on-demand KPI Access (Agentic AI Chatbot):
  • Led end-to-end development of AI chatbot for Pfizer patient data enabling real-time KPI access.
  • Setup MLOps using CI/CD workflows and monitored performances.
  • Saved 400 human working hours, translating to $3M in freed working capital.
  Tech: Data Security, OpenAI, Langchain, FastAPI, Azure Web Apps, Azure DevOps, LLMs, Agentic AI.
• Recommending Next Best Item (Recommendation System):
  • Developed recommender engine to suggest next best drugs to sell for given pharmas.
  • Scaled pipeline to process millions of data records.
  • Enabled extra capital of $2M quarterly based on newly recommended products.
  Tech: Recommendation Engine, Collaborative Filtering, PySpark, Databricks, Snowflake, Keras, PyTorch.
• Out-of-pocket Cost Prediction (Regression Modelling):
  • Developed end-to-end ML system to predict patient out-of-pocket costs for therapy claims.
  • Architected feature engineering pipeline processing 10M+ monthly claims.
  • Engineered two-part models and quantile regression to handle zero-inflated cost distributions.
  Tech: Python, PySpark, Gradient Boosting, Two-Part Models, Quantile Regression, Tweedie Regression, MLflow, Great Expectations.
• Sales Forecasting & Demand Planning (Timeseries Modelling):
  • Developed advanced time series model incorporating historical sales, seasonality and promotional activities.
  • Reduced stockouts by 15% and improved forecast accuracy by 20%, saving $3M.
  Tech: ARIMA, Prophet, LSTM, Pycaret, Greykite, XGBoost, CatBoost, darts, statsmodels, nixtla, TimeGPT, bambi.
• Cross-Functional Dashboard (PowerBI Reports):
  • Created executive dashboards for market share, revenue, and operational KPIs.
  • Collaborated with internal teams and external vendors to unify data pipelines.

Worked as a Data Scientist specializing in Natural Language Processing and data analysis, leading multiple healthcare-related projects and providing end-to-end solutions.

• LLM-Based Text Classification:
  • Fine-tuned BioBERT to classify clinical notes and extract adverse drug events.
  • Enabled real-time alerts for high-risk cases, improving response time by 30%.
  • Implemented Confidence Scoring to evaluate model performance.
  Tech: Transformers, BioBERT, OpenAI, spaCy, sciSpacy, NLP, LLMs.
• Competitor Market Analysis:
  • Conducted market analysis of sales data and presented actionable insights via PowerBI.
  Tech: PowerBI, DAX, Visualization, Business Intelligence, pandas, spark.
• Shipment Delay Anomaly Detection:
  • Built anomaly detection models for logistics delays using PyOD and weather data.
  • Reduced delivery disruptions by 22% through early alerts and root cause analysis.
  Tech: Isolation Forest, Local Outlier Factor, One-Class SVM, PyOD.
• A/B Testing & Web Optimization:
  • Designed and executed A/B tests for a company tools page, leading to 17% increase in user engagement.
  • Performed statistical analysis (Chi-squared test, T-tests) to evaluate metric lift.
  Tech: Experimental Design, Hypothesis Testing, Statistical Significance, Confidence Intervals, Power Analysis, statsmodels, scipy, numpy, pandas.
• Fraud Detection System:
  • Implemented fraud detection model, improving Recall by 15%.
  • Developed custom evaluation metric and used anomaly detection methods as features.
  Tech: XGBoost, LightGBM, scikit-learn, Isolation Forest, Classification, scikit-lego, bambi, statsmodels.